{
  "run": "pilot_max24_curriculum",
  "arm": "max24_curriculum",
  "args": {
    "run_name": "pilot_max24_curriculum",
    "arm": "max24_curriculum",
    "model_id": "Qwen/Qwen3-4B",
    "seed": 123,
    "modulus": 97,
    "stage_max_steps": "24,24,24",
    "stage_min_lengths": "1,1,8",
    "stage_train_max_lengths": "8,16,24",
    "stage_steps": "30,15,30",
    "train_examples": 96,
    "train_template_mode": "mixed",
    "eval_lengths": "8,16,24",
    "eval_examples": 8,
    "paired_eval_pairs": 4,
    "batch_size": 4,
    "grad_accum": 1,
    "eval_batch_size": 4,
    "max_length": 2048,
    "head_width": 256,
    "compiler_layers": 1,
    "compiler_heads": 4,
    "compiler_dropout": 0.05,
    "expansion_noise": 0.005,
    "expansion_mode": "copy_last",
    "trace_loss_weight": 1.0,
    "executor_loss_weight": 1.0,
    "state_loss_weight": 1.0,
    "init_trace_loss_weight": 4.0,
    "op_trace_loss_weight": 1.0,
    "arg_trace_loss_weight": 4.0,
    "direct_head_weight": 0.0,
    "lr": 0.0002,
    "weight_decay": 0.0,
    "optimizer": "paged_adamw_8bit",
    "max_grad_norm": 1.0,
    "torch_dtype": "bf16",
    "load_in_4bit": true,
    "device_map": "auto",
    "use_lora": true,
    "gradient_checkpointing": true,
    "lora_r": 8,
    "lora_alpha": 16,
    "lora_dropout": 0.05,
    "lora_target_modules": "all-linear",
    "log_interval": 15,
    "save_checkpoints": false,
    "checkpoint_dir": "",
    "selection_split": "paired_L24",
    "selection_metric": "executor_pair_both_correct"
  },
  "metadata": {
    "python": "3.12.3",
    "platform": "Linux-6.8.0-62-generic-x86_64-with-glibc2.39",
    "torch": "2.8.0+cu128",
    "cuda_available": true,
    "transformers_loader": "AutoModelForCausalLM",
    "peft_installed": true,
    "use_lora": true,
    "lora_r": 8,
    "model_id": "Qwen/Qwen3-4B",
    "gpu_name": "NVIDIA RTX 6000 Ada Generation",
    "gpu_vram_gb": 47.383
  },
  "stages": [
    {
      "name": "stage1_max24",
      "max_steps": 24,
      "train_min_len": 1,
      "train_max_len": 8,
      "steps": 30
    },
    {
      "name": "stage2_max24",
      "max_steps": 24,
      "train_min_len": 1,
      "train_max_len": 16,
      "steps": 15
    },
    {
      "name": "stage3_max24",
      "max_steps": 24,
      "train_min_len": 8,
      "train_max_len": 24,
      "steps": 30
    }
  ],
  "train_log": [
    {
      "run": "pilot_max24_curriculum",
      "arm": "max24_curriculum",
      "stage": "stage1_max24",
      "stage_idx": 1,
      "local_step": 1,
      "global_step": 1,
      "compiler_max_steps": 24,
      "elapsed_sec": 2.21,
      "expansion_event": "init",
      "init_loss": 4.543251037597656,
      "op_loss": 1.084272027015686,
      "arg_loss": 4.608922004699707,
      "state_loss": 4.579263210296631,
      "state_train_accuracy": 0.0,
      "executor_loss": 4.579021453857422,
      "loss": 46.85124588012695
    },
    {
      "run": "pilot_max24_curriculum",
      "arm": "max24_curriculum",
      "stage": "stage1_max24",
      "stage_idx": 1,
      "local_step": 15,
      "global_step": 15,
      "compiler_max_steps": 24,
      "elapsed_sec": 20.401,
      "expansion_event": "init",
      "init_loss": 4.299290657043457,
      "op_loss": 1.0976250171661377,
      "arg_loss": 4.091684818267822,
      "state_loss": 4.5749030113220215,
      "state_train_accuracy": 0.0,
      "executor_loss": 4.5787458419799805,
      "loss": 43.81517791748047
    },
    {
      "run": "pilot_max24_curriculum",
      "arm": "max24_curriculum",
      "stage": "stage1_max24",
      "stage_idx": 1,
      "local_step": 30,
      "global_step": 30,
      "compiler_max_steps": 24,
      "elapsed_sec": 39.968,
      "expansion_event": "init",
      "init_loss": 4.334928512573242,
      "op_loss": 1.085025668144226,
      "arg_loss": 3.4559853076934814,
      "state_loss": 4.5748395919799805,
      "state_train_accuracy": 0.0,
      "executor_loss": 4.571478843688965,
      "loss": 41.39500045776367
    },
    {
      "run": "pilot_max24_curriculum",
      "arm": "max24_curriculum",
      "stage": "stage2_max24",
      "stage_idx": 2,
      "local_step": 1,
      "global_step": 31,
      "compiler_max_steps": 24,
      "elapsed_sec": 56.897,
      "expansion_event": "same",
      "init_loss": 5.015676975250244,
      "op_loss": 1.0450990200042725,
      "arg_loss": 3.7137184143066406,
      "state_loss": 4.574330806732178,
      "state_train_accuracy": 0.0,
      "executor_loss": 4.576694488525391,
      "loss": 45.113704681396484
    },
    {
      "run": "pilot_max24_curriculum",
      "arm": "max24_curriculum",
      "stage": "stage2_max24",
      "stage_idx": 2,
      "local_step": 15,
      "global_step": 45,
      "compiler_max_steps": 24,
      "elapsed_sec": 76.847,
      "expansion_event": "same",
      "init_loss": 4.588095664978027,
      "op_loss": 1.1084710359573364,
      "arg_loss": 3.8634746074676514,
      "state_loss": 4.572778701782227,
      "state_train_accuracy": 0.032258063554763794,
      "executor_loss": 4.576838493347168,
      "loss": 44.064369201660156
    },
    {
      "run": "pilot_max24_curriculum",
      "arm": "max24_curriculum",
      "stage": "stage3_max24",
      "stage_idx": 3,
      "local_step": 1,
      "global_step": 46,
      "compiler_max_steps": 24,
      "elapsed_sec": 94.304,
      "expansion_event": "same",
      "init_loss": 4.643162727355957,
      "op_loss": 1.1196479797363281,
      "arg_loss": 3.979353427886963,
      "state_loss": 4.577249050140381,
      "state_train_accuracy": 0.0,
      "executor_loss": 4.576903820037842,
      "loss": 44.76386260986328
    },
    {
      "run": "pilot_max24_curriculum",
      "arm": "max24_curriculum",
      "stage": "stage3_max24",
      "stage_idx": 3,
      "local_step": 15,
      "global_step": 60,
      "compiler_max_steps": 24,
      "elapsed_sec": 117.123,
      "expansion_event": "same",
      "init_loss": 4.419005393981934,
      "op_loss": 1.0985978841781616,
      "arg_loss": 3.6761491298675537,
      "state_loss": 4.575003623962402,
      "state_train_accuracy": 0.01587301678955555,
      "executor_loss": 4.575610637664795,
      "loss": 42.62983322143555
    },
    {
      "run": "pilot_max24_curriculum",
      "arm": "max24_curriculum",
      "stage": "stage3_max24",
      "stage_idx": 3,
      "local_step": 30,
      "global_step": 75,
      "compiler_max_steps": 24,
      "elapsed_sec": 141.515,
      "expansion_event": "same",
      "init_loss": 4.656349182128906,
      "op_loss": 1.1077287197113037,
      "arg_loss": 3.673241138458252,
      "state_loss": 4.576873302459717,
      "state_train_accuracy": 0.0,
      "executor_loss": 4.575063228607178,
      "loss": 43.578025817871094
    }
  ],
  "metrics": [
    {
      "stage": "stage1_max24",
      "split": "standard_L8",
      "n": 8,
      "max_steps": 24,
      "executor_accuracy": 0.0,
      "executor_target_mass": 0.010296466760337353,
      "direct_accuracy": NaN,
      "init_accuracy": 0.0,
      "op_accuracy": 0.328125,
      "arg_accuracy": 0.0625,
      "program_exact": 0.0,
      "state_accuracy": 0.015625,
      "state_all_exact": 0.0,
      "state_prefix_fraction": 0.0,
      "state_mean_correct_prefix": 0.0,
      "run": "pilot_max24_curriculum",
      "arm": "max24_curriculum",
      "global_step": 30,
      "stage_idx": 1
    },
    {
      "stage": "stage1_max24",
      "split": "paraphrase_L8",
      "n": 8,
      "max_steps": 24,
      "executor_accuracy": 0.0,
      "executor_target_mass": 0.010294683277606964,
      "direct_accuracy": NaN,
      "init_accuracy": 0.0,
      "op_accuracy": 0.375,
      "arg_accuracy": 0.09375,
      "program_exact": 0.0,
      "state_accuracy": 0.03125,
      "state_all_exact": 0.0,
      "state_prefix_fraction": 0.0,
      "state_mean_correct_prefix": 0.0,
      "run": "pilot_max24_curriculum",
      "arm": "max24_curriculum",
      "global_step": 30,
      "stage_idx": 1
    },
    {
      "stage": "stage1_max24",
      "split": "heldout_L8",
      "n": 8,
      "max_steps": 24,
      "executor_accuracy": 0.0,
      "executor_target_mass": 0.010295983869582415,
      "direct_accuracy": NaN,
      "init_accuracy": 0.0,
      "op_accuracy": 0.34375,
      "arg_accuracy": 0.015625,
      "program_exact": 0.0,
      "state_accuracy": 0.0,
      "state_all_exact": 0.0,
      "state_prefix_fraction": 0.0,
      "state_mean_correct_prefix": 0.0,
      "run": "pilot_max24_curriculum",
      "arm": "max24_curriculum",
      "global_step": 30,
      "stage_idx": 1
    },
    {
      "stage": "stage1_max24",
      "split": "paired_L8",
      "n": 8,
      "max_steps": 24,
      "executor_accuracy": 0.125,
      "executor_target_mass": 0.010303685441613197,
      "direct_accuracy": NaN,
      "init_accuracy": 0.0,
      "op_accuracy": 0.3125,
      "arg_accuracy": 0.03125,
      "program_exact": 0.0,
      "state_accuracy": 0.015625,
      "state_all_exact": 0.0,
      "state_prefix_fraction": 0.0,
      "state_mean_correct_prefix": 0.0,
      "executor_pair_answer_consistency": 0.0,
      "executor_pair_both_correct": 0.0,
      "compiler_pair_program_consistency": 0.0,
      "compiler_pair_state_consistency": 0.0,
      "run": "pilot_max24_curriculum",
      "arm": "max24_curriculum",
      "global_step": 30,
      "stage_idx": 1
    },
    {
      "stage": "stage1_max24",
      "split": "paired_heldout_L8",
      "n": 8,
      "max_steps": 24,
      "executor_accuracy": 0.0,
      "executor_target_mass": 0.010293223895132542,
      "direct_accuracy": NaN,
      "init_accuracy": 0.0,
      "op_accuracy": 0.4375,
      "arg_accuracy": 0.015625,
      "program_exact": 0.0,
      "state_accuracy": 0.015625,
      "state_all_exact": 0.0,
      "state_prefix_fraction": 0.0,
      "state_mean_correct_prefix": 0.0,
      "executor_pair_answer_consistency": 0.0,
      "executor_pair_both_correct": 0.0,
      "compiler_pair_program_consistency": 0.0,
      "compiler_pair_state_consistency": 0.0,
      "run": "pilot_max24_curriculum",
      "arm": "max24_curriculum",
      "global_step": 30,
      "stage_idx": 1
    },
    {
      "stage": "stage1_max24",
      "split": "standard_L16",
      "n": 8,
      "max_steps": 24,
      "executor_accuracy": 0.0,
      "executor_target_mass": 0.010294985957443714,
      "direct_accuracy": NaN,
      "init_accuracy": 0.0,
      "op_accuracy": 0.3359375,
      "arg_accuracy": 0.0859375,
      "program_exact": 0.0,
      "state_accuracy": 0.015625,
      "state_all_exact": 0.0,
      "state_prefix_fraction": 0.0078125,
      "state_mean_correct_prefix": 0.125,
      "run": "pilot_max24_curriculum",
      "arm": "max24_curriculum",
      "global_step": 30,
      "stage_idx": 1
    },
    {
      "stage": "stage1_max24",
      "split": "paraphrase_L16",
      "n": 8,
      "max_steps": 24,
      "executor_accuracy": 0.0,
      "executor_target_mass": 0.010297542437911034,
      "direct_accuracy": NaN,
      "init_accuracy": 0.0,
      "op_accuracy": 0.2734375,
      "arg_accuracy": 0.03125,
      "program_exact": 0.0,
      "state_accuracy": 0.0,
      "state_all_exact": 0.0,
      "state_prefix_fraction": 0.0,
      "state_mean_correct_prefix": 0.0,
      "run": "pilot_max24_curriculum",
      "arm": "max24_curriculum",
      "global_step": 30,
      "stage_idx": 1
    },
    {
      "stage": "stage1_max24",
      "split": "heldout_L16",
      "n": 8,
      "max_steps": 24,
      "executor_accuracy": 0.0,
      "executor_target_mass": 0.010295731946825981,
      "direct_accuracy": NaN,
      "init_accuracy": 0.0,
      "op_accuracy": 0.3984375,
      "arg_accuracy": 0.0546875,
      "program_exact": 0.0,
      "state_accuracy": 0.0078125,
      "state_all_exact": 0.0,
      "state_prefix_fraction": 0.0,
      "state_mean_correct_prefix": 0.0,
      "run": "pilot_max24_curriculum",
      "arm": "max24_curriculum",
      "global_step": 30,
      "stage_idx": 1
    },
    {
      "stage": "stage1_max24",
      "split": "paired_L16",
      "n": 8,
      "max_steps": 24,
      "executor_accuracy": 0.0,
      "executor_target_mass": 0.010291510727256536,
      "direct_accuracy": NaN,
      "init_accuracy": 0.0,
      "op_accuracy": 0.203125,
      "arg_accuracy": 0.0703125,
      "program_exact": 0.0,
      "state_accuracy": 0.0,
      "state_all_exact": 0.0,
      "state_prefix_fraction": 0.0,
      "state_mean_correct_prefix": 0.0,
      "executor_pair_answer_consistency": 0.0,
      "executor_pair_both_correct": 0.0,
      "compiler_pair_program_consistency": 0.0,
      "compiler_pair_state_consistency": 0.0,
      "run": "pilot_max24_curriculum",
      "arm": "max24_curriculum",
      "global_step": 30,
      "stage_idx": 1
    },
    {
      "stage": "stage1_max24",
      "split": "paired_heldout_L16",
      "n": 8,
      "max_steps": 24,
      "executor_accuracy": 0.0,
      "executor_target_mass": 0.010303504299372435,
      "direct_accuracy": NaN,
      "init_accuracy": 0.0,
      "op_accuracy": 0.359375,
      "arg_accuracy": 0.03125,
      "program_exact": 0.0,
      "state_accuracy": 0.015625,
      "state_all_exact": 0.0,
      "state_prefix_fraction": 0.0,
      "state_mean_correct_prefix": 0.0,
      "executor_pair_answer_consistency": 0.25,
      "executor_pair_both_correct": 0.0,
      "compiler_pair_program_consistency": 0.0,
      "compiler_pair_state_consistency": 0.0,
      "run": "pilot_max24_curriculum",
      "arm": "max24_curriculum",
      "global_step": 30,
      "stage_idx": 1
    },
    {
      "stage": "stage1_max24",
      "split": "standard_L24",
      "n": 8,
      "max_steps": 24,
      "executor_accuracy": 0.0,
      "executor_target_mass": 0.010295350104570389,
      "direct_accuracy": NaN,
      "init_accuracy": 0.0,
      "op_accuracy": 0.359375,
      "arg_accuracy": 0.057291666666666664,
      "program_exact": 0.0,
      "state_accuracy": 0.020833333333333332,
      "state_all_exact": 0.0,
      "state_prefix_fraction": 0.0,
      "state_mean_correct_prefix": 0.0,
      "run": "pilot_max24_curriculum",
      "arm": "max24_curriculum",
      "global_step": 30,
      "stage_idx": 1
    },
    {
      "stage": "stage1_max24",
      "split": "paraphrase_L24",
      "n": 8,
      "max_steps": 24,
      "executor_accuracy": 0.0,
      "executor_target_mass": 0.010295339860022068,
      "direct_accuracy": NaN,
      "init_accuracy": 0.0,
      "op_accuracy": 0.2760416666666667,
      "arg_accuracy": 0.046875,
      "program_exact": 0.0,
      "state_accuracy": 0.010416666666666666,
      "state_all_exact": 0.0,
      "state_prefix_fraction": 0.0,
      "state_mean_correct_prefix": 0.0,
      "run": "pilot_max24_curriculum",
      "arm": "max24_curriculum",
      "global_step": 30,
      "stage_idx": 1
    },
    {
      "stage": "stage1_max24",
      "split": "heldout_L24",
      "n": 8,
      "max_steps": 24,
      "executor_accuracy": 0.0,
      "executor_target_mass": 0.010294530540704727,
      "direct_accuracy": NaN,
      "init_accuracy": 0.0,
      "op_accuracy": 0.2916666666666667,
      "arg_accuracy": 0.052083333333333336,
      "program_exact": 0.0,
      "state_accuracy": 0.010416666666666666,
      "state_all_exact": 0.0,
      "state_prefix_fraction": 0.0,
      "state_mean_correct_prefix": 0.0,
      "run": "pilot_max24_curriculum",
      "arm": "max24_curriculum",
      "global_step": 30,
      "stage_idx": 1
    },
    {
      "stage": "stage1_max24",
      "split": "paired_L24",
      "n": 8,
      "max_steps": 24,
      "executor_accuracy": 0.0,
      "executor_target_mass": 0.01029663672670722,
      "direct_accuracy": NaN,
      "init_accuracy": 0.0,
      "op_accuracy": 0.375,
      "arg_accuracy": 0.03125,
      "program_exact": 0.0,
      "state_accuracy": 0.010416666666666666,
      "state_all_exact": 0.0,
      "state_prefix_fraction": 0.0,
      "state_mean_correct_prefix": 0.0,
      "executor_pair_answer_consistency": 0.0,
      "executor_pair_both_correct": 0.0,
      "compiler_pair_program_consistency": 0.0,
      "compiler_pair_state_consistency": 0.0,
      "run": "pilot_max24_curriculum",
      "arm": "max24_curriculum",
      "global_step": 30,
      "stage_idx": 1
    },
    {
      "stage": "stage1_max24",
      "split": "paired_heldout_L24",
      "n": 8,
      "max_steps": 24,
      "executor_accuracy": 0.0,
      "executor_target_mass": 0.010296373162418604,
      "direct_accuracy": NaN,
      "init_accuracy": 0.0,
      "op_accuracy": 0.3229166666666667,
      "arg_accuracy": 0.06770833333333333,
      "program_exact": 0.0,
      "state_accuracy": 0.0,
      "state_all_exact": 0.0,
      "state_prefix_fraction": 0.0,
      "state_mean_correct_prefix": 0.0,
      "executor_pair_answer_consistency": 0.0,
      "executor_pair_both_correct": 0.0,
      "compiler_pair_program_consistency": 0.0,
      "compiler_pair_state_consistency": 0.0,
      "run": "pilot_max24_curriculum",
      "arm": "max24_curriculum",
      "global_step": 30,
      "stage_idx": 1
    },
    {
      "stage": "stage2_max24",
      "split": "standard_L8",
      "n": 8,
      "max_steps": 24,
      "executor_accuracy": 0.0,
      "executor_target_mass": 0.01050786767154932,
      "direct_accuracy": NaN,
      "init_accuracy": 0.0,
      "op_accuracy": 0.296875,
      "arg_accuracy": 0.078125,
      "program_exact": 0.0,
      "state_accuracy": 0.0,
      "state_all_exact": 0.0,
      "state_prefix_fraction": 0.0,
      "state_mean_correct_prefix": 0.0,
      "run": "pilot_max24_curriculum",
      "arm": "max24_curriculum",
      "global_step": 45,
      "stage_idx": 2
    },
    {
      "stage": "stage2_max24",
      "split": "paraphrase_L8",
      "n": 8,
      "max_steps": 24,
      "executor_accuracy": 0.0,
      "executor_target_mass": 0.010291456244885921,
      "direct_accuracy": NaN,
      "init_accuracy": 0.0,
      "op_accuracy": 0.34375,
      "arg_accuracy": 0.09375,
      "program_exact": 0.0,
      "state_accuracy": 0.0,
      "state_all_exact": 0.0,
      "state_prefix_fraction": 0.0,
      "state_mean_correct_prefix": 0.0,
      "run": "pilot_max24_curriculum",
      "arm": "max24_curriculum",
      "global_step": 45,
      "stage_idx": 2
    },
    {
      "stage": "stage2_max24",
      "split": "heldout_L8",
      "n": 8,
      "max_steps": 24,
      "executor_accuracy": 0.0,
      "executor_target_mass": 0.010288253892213106,
      "direct_accuracy": NaN,
      "init_accuracy": 0.0,
      "op_accuracy": 0.328125,
      "arg_accuracy": 0.046875,
      "program_exact": 0.0,
      "state_accuracy": 0.0,
      "state_all_exact": 0.0,
      "state_prefix_fraction": 0.0,
      "state_mean_correct_prefix": 0.0,
      "run": "pilot_max24_curriculum",
      "arm": "max24_curriculum",
      "global_step": 45,
      "stage_idx": 2
    },
    {
      "stage": "stage2_max24",
      "split": "paired_L8",
      "n": 8,
      "max_steps": 24,
      "executor_accuracy": 0.0,
      "executor_target_mass": 0.010289593134075403,
      "direct_accuracy": NaN,
      "init_accuracy": 0.0,
      "op_accuracy": 0.3125,
      "arg_accuracy": 0.0625,
      "program_exact": 0.0,
      "state_accuracy": 0.0,
      "state_all_exact": 0.0,
      "state_prefix_fraction": 0.0,
      "state_mean_correct_prefix": 0.0,
      "executor_pair_answer_consistency": 0.25,
      "executor_pair_both_correct": 0.0,
      "compiler_pair_program_consistency": 0.0,
      "compiler_pair_state_consistency": 0.0,
      "run": "pilot_max24_curriculum",
      "arm": "max24_curriculum",
      "global_step": 45,
      "stage_idx": 2
    },
    {
      "stage": "stage2_max24",
      "split": "paired_heldout_L8",
      "n": 8,
      "max_steps": 24,
      "executor_accuracy": 0.125,
      "executor_target_mass": 0.010286081116646528,
      "direct_accuracy": NaN,
      "init_accuracy": 0.0,
      "op_accuracy": 0.28125,
      "arg_accuracy": 0.078125,
      "program_exact": 0.0,
      "state_accuracy": 0.0625,
      "state_all_exact": 0.0,
      "state_prefix_fraction": 0.0,
      "state_mean_correct_prefix": 0.0,
      "executor_pair_answer_consistency": 0.0,
      "executor_pair_both_correct": 0.0,
      "compiler_pair_program_consistency": 0.0,
      "compiler_pair_state_consistency": 0.0,
      "run": "pilot_max24_curriculum",
      "arm": "max24_curriculum",
      "global_step": 45,
      "stage_idx": 2
    },
    {
      "stage": "stage2_max24",
      "split": "standard_L16",
      "n": 8,
      "max_steps": 24,
      "executor_accuracy": 0.0,
      "executor_target_mass": 0.010290623176842928,
      "direct_accuracy": NaN,
      "init_accuracy": 0.0,
      "op_accuracy": 0.296875,
      "arg_accuracy": 0.015625,
      "program_exact": 0.0,
      "state_accuracy": 0.0078125,
      "state_all_exact": 0.0,
      "state_prefix_fraction": 0.0,
      "state_mean_correct_prefix": 0.0,
      "run": "pilot_max24_curriculum",
      "arm": "max24_curriculum",
      "global_step": 45,
      "stage_idx": 2
    },
    {
      "stage": "stage2_max24",
      "split": "paraphrase_L16",
      "n": 8,
      "max_steps": 24,
      "executor_accuracy": 0.0,
      "executor_target_mass": 0.010285988450050354,
      "direct_accuracy": NaN,
      "init_accuracy": 0.0,
      "op_accuracy": 0.390625,
      "arg_accuracy": 0.0859375,
      "program_exact": 0.0,
      "state_accuracy": 0.0078125,
      "state_all_exact": 0.0,
      "state_prefix_fraction": 0.0,
      "state_mean_correct_prefix": 0.0,
      "run": "pilot_max24_curriculum",
      "arm": "max24_curriculum",
      "global_step": 45,
      "stage_idx": 2
    },
    {
      "stage": "stage2_max24",
      "split": "heldout_L16",
      "n": 8,
      "max_steps": 24,
      "executor_accuracy": 0.0,
      "executor_target_mass": 0.01028686249628663,
      "direct_accuracy": NaN,
      "init_accuracy": 0.0,
      "op_accuracy": 0.3203125,
      "arg_accuracy": 0.0390625,
      "program_exact": 0.0,
      "state_accuracy": 0.015625,
      "state_all_exact": 0.0,
      "state_prefix_fraction": 0.0,
      "state_mean_correct_prefix": 0.0,
      "run": "pilot_max24_curriculum",
      "arm": "max24_curriculum",
      "global_step": 45,
      "stage_idx": 2
    },
    {
      "stage": "stage2_max24",
      "split": "paired_L16",
      "n": 8,
      "max_steps": 24,
      "executor_accuracy": 0.0,
      "executor_target_mass": 0.010291657410562038,
      "direct_accuracy": NaN,
      "init_accuracy": 0.0,
      "op_accuracy": 0.375,
      "arg_accuracy": 0.1015625,
      "program_exact": 0.0,
      "state_accuracy": 0.03125,
      "state_all_exact": 0.0,
      "state_prefix_fraction": 0.0,
      "state_mean_correct_prefix": 0.0,
      "executor_pair_answer_consistency": 0.0,
      "executor_pair_both_correct": 0.0,
      "compiler_pair_program_consistency": 0.0,
      "compiler_pair_state_consistency": 0.0,
      "run": "pilot_max24_curriculum",
      "arm": "max24_curriculum",
      "global_step": 45,
      "stage_idx": 2
    },
    {
      "stage": "stage2_max24",
      "split": "paired_heldout_L16",
      "n": 8,
      "max_steps": 24,
      "executor_accuracy": 0.0,
      "executor_target_mass": 0.01028924947604537,
      "direct_accuracy": NaN,
      "init_accuracy": 0.0,
      "op_accuracy": 0.375,
      "arg_accuracy": 0.078125,
      "program_exact": 0.0,
      "state_accuracy": 0.0078125,
      "state_all_exact": 0.0,
      "state_prefix_fraction": 0.0,
      "state_mean_correct_prefix": 0.0,
      "executor_pair_answer_consistency": 0.0,
      "executor_pair_both_correct": 0.0,
      "compiler_pair_program_consistency": 0.0,
      "compiler_pair_state_consistency": 0.0,
      "run": "pilot_max24_curriculum",
      "arm": "max24_curriculum",
      "global_step": 45,
      "stage_idx": 2
    },
    {
      "stage": "stage2_max24",
      "split": "standard_L24",
      "n": 8,
      "max_steps": 24,
      "executor_accuracy": 0.125,
      "executor_target_mass": 0.010294968262314796,
      "direct_accuracy": NaN,
      "init_accuracy": 0.0,
      "op_accuracy": 0.3125,
      "arg_accuracy": 0.06770833333333333,
      "program_exact": 0.0,
      "state_accuracy": 0.015625,
      "state_all_exact": 0.0,
      "state_prefix_fraction": 0.0,
      "state_mean_correct_prefix": 0.0,
      "run": "pilot_max24_curriculum",
      "arm": "max24_curriculum",
      "global_step": 45,
      "stage_idx": 2
    },
    {
      "stage": "stage2_max24",
      "split": "paraphrase_L24",
      "n": 8,
      "max_steps": 24,
      "executor_accuracy": 0.0,
      "executor_target_mass": 0.010296504478901625,
      "direct_accuracy": NaN,
      "init_accuracy": 0.0,
      "op_accuracy": 0.3697916666666667,
      "arg_accuracy": 0.0625,
      "program_exact": 0.0,
      "state_accuracy": 0.020833333333333332,
      "state_all_exact": 0.0,
      "state_prefix_fraction": 0.0,
      "state_mean_correct_prefix": 0.0,
      "run": "pilot_max24_curriculum",
      "arm": "max24_curriculum",
      "global_step": 45,
      "stage_idx": 2
    },
    {
      "stage": "stage2_max24",
      "split": "heldout_L24",
      "n": 8,
      "max_steps": 24,
      "executor_accuracy": 0.0,
      "executor_target_mass": 0.01028955914080143,
      "direct_accuracy": NaN,
      "init_accuracy": 0.0,
      "op_accuracy": 0.3541666666666667,
      "arg_accuracy": 0.057291666666666664,
      "program_exact": 0.0,
      "state_accuracy": 0.0,
      "state_all_exact": 0.0,
      "state_prefix_fraction": 0.0,
      "state_mean_correct_prefix": 0.0,
      "run": "pilot_max24_curriculum",
      "arm": "max24_curriculum",
      "global_step": 45,
      "stage_idx": 2
    },
    {
      "stage": "stage2_max24",
      "split": "paired_L24",
      "n": 8,
      "max_steps": 24,
      "executor_accuracy": 0.0,
      "executor_target_mass": 0.010296678636223078,
      "direct_accuracy": NaN,
      "init_accuracy": 0.0,
      "op_accuracy": 0.375,
      "arg_accuracy": 0.041666666666666664,
      "program_exact": 0.0,
      "state_accuracy": 0.010416666666666666,
      "state_all_exact": 0.0,
      "state_prefix_fraction": 0.0,
      "state_mean_correct_prefix": 0.0,
      "executor_pair_answer_consistency": 0.0,
      "executor_pair_both_correct": 0.0,
      "compiler_pair_program_consistency": 0.0,
      "compiler_pair_state_consistency": 0.0,
      "run": "pilot_max24_curriculum",
      "arm": "max24_curriculum",
      "global_step": 45,
      "stage_idx": 2
    },
    {
      "stage": "stage2_max24",
      "split": "paired_heldout_L24",
      "n": 8,
      "max_steps": 24,
      "executor_accuracy": 0.0,
      "executor_target_mass": 0.01029588794335723,
      "direct_accuracy": NaN,
      "init_accuracy": 0.0,
      "op_accuracy": 0.296875,
      "arg_accuracy": 0.010416666666666666,
      "program_exact": 0.0,
      "state_accuracy": 0.0,
      "state_all_exact": 0.0,
      "state_prefix_fraction": 0.0,
      "state_mean_correct_prefix": 0.0,
      "executor_pair_answer_consistency": 0.0,
      "executor_pair_both_correct": 0.0,
      "compiler_pair_program_consistency": 0.0,
      "compiler_pair_state_consistency": 0.0,
      "run": "pilot_max24_curriculum",
      "arm": "max24_curriculum",
      "global_step": 45,
      "stage_idx": 2
    },
    {
      "stage": "stage3_max24",
      "split": "standard_L8",
      "n": 8,
      "max_steps": 24,
      "executor_accuracy": 0.0,
      "executor_target_mass": 0.01030506705865264,
      "direct_accuracy": NaN,
      "init_accuracy": 0.0,
      "op_accuracy": 0.28125,
      "arg_accuracy": 0.0625,
      "program_exact": 0.0,
      "state_accuracy": 0.0,
      "state_all_exact": 0.0,
      "state_prefix_fraction": 0.0,
      "state_mean_correct_prefix": 0.0,
      "run": "pilot_max24_curriculum",
      "arm": "max24_curriculum",
      "global_step": 75,
      "stage_idx": 3
    },
    {
      "stage": "stage3_max24",
      "split": "paraphrase_L8",
      "n": 8,
      "max_steps": 24,
      "executor_accuracy": 0.0,
      "executor_target_mass": 0.0103058572858572,
      "direct_accuracy": NaN,
      "init_accuracy": 0.0,
      "op_accuracy": 0.4375,
      "arg_accuracy": 0.046875,
      "program_exact": 0.0,
      "state_accuracy": 0.0,
      "state_all_exact": 0.0,
      "state_prefix_fraction": 0.0,
      "state_mean_correct_prefix": 0.0,
      "run": "pilot_max24_curriculum",
      "arm": "max24_curriculum",
      "global_step": 75,
      "stage_idx": 3
    },
    {
      "stage": "stage3_max24",
      "split": "heldout_L8",
      "n": 8,
      "max_steps": 24,
      "executor_accuracy": 0.0,
      "executor_target_mass": 0.010306863114237785,
      "direct_accuracy": NaN,
      "init_accuracy": 0.125,
      "op_accuracy": 0.28125,
      "arg_accuracy": 0.0625,
      "program_exact": 0.0,
      "state_accuracy": 0.0,
      "state_all_exact": 0.0,
      "state_prefix_fraction": 0.0,
      "state_mean_correct_prefix": 0.0,
      "run": "pilot_max24_curriculum",
      "arm": "max24_curriculum",
      "global_step": 75,
      "stage_idx": 3
    },
    {
      "stage": "stage3_max24",
      "split": "paired_L8",
      "n": 8,
      "max_steps": 24,
      "executor_accuracy": 0.0,
      "executor_target_mass": 0.010307054966688156,
      "direct_accuracy": NaN,
      "init_accuracy": 0.0,
      "op_accuracy": 0.328125,
      "arg_accuracy": 0.03125,
      "program_exact": 0.0,
      "state_accuracy": 0.0,
      "state_all_exact": 0.0,
      "state_prefix_fraction": 0.0,
      "state_mean_correct_prefix": 0.0,
      "executor_pair_answer_consistency": 0.75,
      "executor_pair_both_correct": 0.0,
      "compiler_pair_program_consistency": 0.75,
      "compiler_pair_state_consistency": 0.75,
      "run": "pilot_max24_curriculum",
      "arm": "max24_curriculum",
      "global_step": 75,
      "stage_idx": 3
    },
    {
      "stage": "stage3_max24",
      "split": "paired_heldout_L8",
      "n": 8,
      "max_steps": 24,
      "executor_accuracy": 0.0,
      "executor_target_mass": 0.010306525975465775,
      "direct_accuracy": NaN,
      "init_accuracy": 0.0,
      "op_accuracy": 0.46875,
      "arg_accuracy": 0.0,
      "program_exact": 0.0,
      "state_accuracy": 0.0,
      "state_all_exact": 0.0,
      "state_prefix_fraction": 0.0,
      "state_mean_correct_prefix": 0.0,
      "executor_pair_answer_consistency": 0.25,
      "executor_pair_both_correct": 0.0,
      "compiler_pair_program_consistency": 0.25,
      "compiler_pair_state_consistency": 0.25,
      "run": "pilot_max24_curriculum",
      "arm": "max24_curriculum",
      "global_step": 75,
      "stage_idx": 3
    },
    {
      "stage": "stage3_max24",
      "split": "standard_L16",
      "n": 8,
      "max_steps": 24,
      "executor_accuracy": 0.0,
      "executor_target_mass": 0.010307608637958765,
      "direct_accuracy": NaN,
      "init_accuracy": 0.0,
      "op_accuracy": 0.296875,
      "arg_accuracy": 0.0390625,
      "program_exact": 0.0,
      "state_accuracy": 0.0,
      "state_all_exact": 0.0,
      "state_prefix_fraction": 0.0,
      "state_mean_correct_prefix": 0.0,
      "run": "pilot_max24_curriculum",
      "arm": "max24_curriculum",
      "global_step": 75,
      "stage_idx": 3
    },
    {
      "stage": "stage3_max24",
      "split": "paraphrase_L16",
      "n": 8,
      "max_steps": 24,
      "executor_accuracy": 0.0,
      "executor_target_mass": 0.010306145995855331,
      "direct_accuracy": NaN,
      "init_accuracy": 0.0,
      "op_accuracy": 0.3359375,
      "arg_accuracy": 0.0390625,
      "program_exact": 0.0,
      "state_accuracy": 0.0078125,
      "state_all_exact": 0.0,
      "state_prefix_fraction": 0.0,
      "state_mean_correct_prefix": 0.0,
      "run": "pilot_max24_curriculum",
      "arm": "max24_curriculum",
      "global_step": 75,
      "stage_idx": 3
    },
    {
      "stage": "stage3_max24",
      "split": "heldout_L16",
      "n": 8,
      "max_steps": 24,
      "executor_accuracy": 0.0,
      "executor_target_mass": 0.010306648444384336,
      "direct_accuracy": NaN,
      "init_accuracy": 0.0,
      "op_accuracy": 0.328125,
      "arg_accuracy": 0.0546875,
      "program_exact": 0.0,
      "state_accuracy": 0.0,
      "state_all_exact": 0.0,
      "state_prefix_fraction": 0.0,
      "state_mean_correct_prefix": 0.0,
      "run": "pilot_max24_curriculum",
      "arm": "max24_curriculum",
      "global_step": 75,
      "stage_idx": 3
    },
    {
      "stage": "stage3_max24",
      "split": "paired_L16",
      "n": 8,
      "max_steps": 24,
      "executor_accuracy": 0.0,
      "executor_target_mass": 0.010305347852408886,
      "direct_accuracy": NaN,
      "init_accuracy": 0.0,
      "op_accuracy": 0.40625,
      "arg_accuracy": 0.0625,
      "program_exact": 0.0,
      "state_accuracy": 0.015625,
      "state_all_exact": 0.0,
      "state_prefix_fraction": 0.0,
      "state_mean_correct_prefix": 0.0,
      "executor_pair_answer_consistency": 0.0,
      "executor_pair_both_correct": 0.0,
      "compiler_pair_program_consistency": 0.0,
      "compiler_pair_state_consistency": 0.0,
      "run": "pilot_max24_curriculum",
      "arm": "max24_curriculum",
      "global_step": 75,
      "stage_idx": 3
    },
    {
      "stage": "stage3_max24",
      "split": "paired_heldout_L16",
      "n": 8,
      "max_steps": 24,
      "executor_accuracy": 0.0,
      "executor_target_mass": 0.010307649616152048,
      "direct_accuracy": NaN,
      "init_accuracy": 0.0,
      "op_accuracy": 0.359375,
      "arg_accuracy": 0.03125,
      "program_exact": 0.0,
      "state_accuracy": 0.03125,
      "state_all_exact": 0.0,
      "state_prefix_fraction": 0.0,
      "state_mean_correct_prefix": 0.0,
      "executor_pair_answer_consistency": 0.0,
      "executor_pair_both_correct": 0.0,
      "compiler_pair_program_consistency": 0.0,
      "compiler_pair_state_consistency": 0.0,
      "run": "pilot_max24_curriculum",
      "arm": "max24_curriculum",
      "global_step": 75,
      "stage_idx": 3
    },
    {
      "stage": "stage3_max24",
      "split": "standard_L24",
      "n": 8,
      "max_steps": 24,
      "executor_accuracy": 0.0,
      "executor_target_mass": 0.010305903386324644,
      "direct_accuracy": NaN,
      "init_accuracy": 0.0,
      "op_accuracy": 0.3072916666666667,
      "arg_accuracy": 0.0625,
      "program_exact": 0.0,
      "state_accuracy": 0.005208333333333333,
      "state_all_exact": 0.0,
      "state_prefix_fraction": 0.0,
      "state_mean_correct_prefix": 0.0,
      "run": "pilot_max24_curriculum",
      "arm": "max24_curriculum",
      "global_step": 75,
      "stage_idx": 3
    },
    {
      "stage": "stage3_max24",
      "split": "paraphrase_L24",
      "n": 8,
      "max_steps": 24,
      "executor_accuracy": 0.0,
      "executor_target_mass": 0.010343425441533327,
      "direct_accuracy": NaN,
      "init_accuracy": 0.125,
      "op_accuracy": 0.2552083333333333,
      "arg_accuracy": 0.052083333333333336,
      "program_exact": 0.0,
      "state_accuracy": 0.010416666666666666,
      "state_all_exact": 0.0,
      "state_prefix_fraction": 0.0,
      "state_mean_correct_prefix": 0.0,
      "run": "pilot_max24_curriculum",
      "arm": "max24_curriculum",
      "global_step": 75,
      "stage_idx": 3
    },
    {
      "stage": "stage3_max24",
      "split": "heldout_L24",
      "n": 8,
      "max_steps": 24,
      "executor_accuracy": 0.0,
      "executor_target_mass": 0.010306162759661674,
      "direct_accuracy": NaN,
      "init_accuracy": 0.0,
      "op_accuracy": 0.296875,
      "arg_accuracy": 0.046875,
      "program_exact": 0.0,
      "state_accuracy": 0.005208333333333333,
      "state_all_exact": 0.0,
      "state_prefix_fraction": 0.0,
      "state_mean_correct_prefix": 0.0,
      "run": "pilot_max24_curriculum",
      "arm": "max24_curriculum",
      "global_step": 75,
      "stage_idx": 3
    },
    {
      "stage": "stage3_max24",
      "split": "paired_L24",
      "n": 8,
      "max_steps": 24,
      "executor_accuracy": 0.0,
      "executor_target_mass": 0.010305967181921005,
      "direct_accuracy": NaN,
      "init_accuracy": 0.0,
      "op_accuracy": 0.3177083333333333,
      "arg_accuracy": 0.03125,
      "program_exact": 0.0,
      "state_accuracy": 0.010416666666666666,
      "state_all_exact": 0.0,
      "state_prefix_fraction": 0.0,
      "state_mean_correct_prefix": 0.0,
      "executor_pair_answer_consistency": 0.0,
      "executor_pair_both_correct": 0.0,
      "compiler_pair_program_consistency": 0.0,
      "compiler_pair_state_consistency": 0.0,
      "run": "pilot_max24_curriculum",
      "arm": "max24_curriculum",
      "global_step": 75,
      "stage_idx": 3
    },
    {
      "stage": "stage3_max24",
      "split": "paired_heldout_L24",
      "n": 8,
      "max_steps": 24,
      "executor_accuracy": 0.0,
      "executor_target_mass": 0.010305618867278099,
      "direct_accuracy": NaN,
      "init_accuracy": 0.0,
      "op_accuracy": 0.375,
      "arg_accuracy": 0.052083333333333336,
      "program_exact": 0.0,
      "state_accuracy": 0.026041666666666668,
      "state_all_exact": 0.0,
      "state_prefix_fraction": 0.0,
      "state_mean_correct_prefix": 0.0,
      "executor_pair_answer_consistency": 0.5,
      "executor_pair_both_correct": 0.0,
      "compiler_pair_program_consistency": 0.25,
      "compiler_pair_state_consistency": 0.25,
      "run": "pilot_max24_curriculum",
      "arm": "max24_curriculum",
      "global_step": 75,
      "stage_idx": 3
    }
  ],
  "checkpoints": [],
  "elapsed_sec": 156.951
}